src/mesa/drivers/dri/i965/brw_wm.c

   1 /*
   2  Copyright (C) Intel Corp.  2006.  All Rights Reserved.
   3  Intel funded Tungsten Graphics (http://www.tungstengraphics.com) to
   4  develop this 3D driver.
   5
   6  Permission is hereby granted, free of charge, to any person obtaining
   7  a copy of this software and associated documentation files (the
   8  "Software"), to deal in the Software without restriction, including
   9  without limitation the rights to use, copy, modify, merge, publish,
  10  distribute, sublicense, and/or sell copies of the Software, and to
  11  permit persons to whom the Software is furnished to do so, subject to
  12  the following conditions:
  13
  14  The above copyright notice and this permission notice (including the
  15  next paragraph) shall be included in all copies or substantial
  16  portions of the Software.
  17
  18  THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
  19  EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
  20  MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
  21  IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE
  22  LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
  23  OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
  24  WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
  25
  26  **********************************************************************/
  27  /*
  28   * Authors:
  29   *   Keith Whitwell <keith@tungstengraphics.com>
  30   */
  31
  32 #include "main/texformat.h"
  33 #include "brw_context.h"
  34 #include "brw_util.h"
  35 #include "brw_wm.h"
  36 #include "brw_state.h"
  37
  38
  39 /** Return number of src args for given instruction */
  40 GLuint brw_wm_nr_args( GLuint opcode )
  41 {
  42    switch (opcode) {
  43    case WM_PIXELXY:
  44    case WM_CINTERP:
  45    case WM_WPOSXY:
  46       return 1;
  47    case WM_LINTERP:
  48    case WM_DELTAXY:
  49    case WM_PIXELW:
  50       return 2;
  51    case WM_FB_WRITE:
  52    case WM_PINTERP:
  53       return 3;
  54    default:
  55       assert(opcode < MAX_OPCODE);
  56       return _mesa_num_inst_src_regs(opcode);
  57    }
  58 }
  59
  60
  61 GLuint brw_wm_is_scalar_result( GLuint opcode )
  62 {
  63    switch (opcode) {
  64    case OPCODE_COS:
  65    case OPCODE_EX2:
  66    case OPCODE_LG2:
  67    case OPCODE_POW:
  68    case OPCODE_RCP:
  69    case OPCODE_RSQ:
  70    case OPCODE_SIN:
  71    case OPCODE_DP3:
  72    case OPCODE_DP4:
  73    case OPCODE_DPH:
  74    case OPCODE_DST:
  75       return 1;
  76
  77    default:
  78       return 0;
  79    }
  80 }
  81
  82
  83 static void do_wm_prog( struct brw_context *brw,
  84                         struct brw_fragment_program *fp,
  85                         struct brw_wm_prog_key *key)
  86 {
  87    struct brw_wm_compile *c;
  88    const GLuint *program;
  89    GLuint program_size;
  90
  91    c = brw->wm.compile_data;
  92    if (c == NULL) {
  93      brw->wm.compile_data = calloc(1, sizeof(*brw->wm.compile_data));
  94      c = brw->wm.compile_data;
  95    } else {
  96      memset(c, 0, sizeof(*brw->wm.compile_data));
  97    }
  98    memcpy(&c->key, key, sizeof(*key));
  99
 100    c->fp = fp;
 101    c->env_param = brw->intel.ctx.FragmentProgram.Parameters;
 102
 103     brw_init_compile(brw, &c->func);
 104    if (brw_wm_is_glsl(&c->fp->program)) {
 105        brw_wm_glsl_emit(brw, c);
 106    } else {
 107        /* Augment fragment program.  Add instructions for pre- and
 108         * post-fragment-program tasks such as interpolation and fogging.
 109         */
 110        brw_wm_pass_fp(c);
 111
 112        /* Translate to intermediate representation.  Build register usage
 113         * chains.
 114         */
 115        brw_wm_pass0(c);
 116
 117        /* Dead code removal.
 118         */
 119        brw_wm_pass1(c);
 120
 121        /* Register allocation.
 122         */
 123        c->grf_limit = BRW_WM_MAX_GRF/2;
 124
 125        brw_wm_pass2(c);
 126
 127        c->prog_data.total_grf = c->max_wm_grf;
 128        if (c->last_scratch) {
 129            c->prog_data.total_scratch =
 130                c->last_scratch + 0x40;
 131        } else {
 132            c->prog_data.total_scratch = 0;
 133        }
 134
 135        /* Emit GEN4 code.
 136         */
 137        brw_wm_emit(c);
 138    }
 139    if (INTEL_DEBUG & DEBUG_WM)
 140       fprintf(stderr, "\n");
 141
 142    /* get the program
 143     */
 144    program = brw_get_program(&c->func, &program_size);
 145
 146    dri_bo_unreference(brw->wm.prog_bo);
 147    brw->wm.prog_bo = brw_upload_cache( &brw->cache, BRW_WM_PROG,
 148                                        &c->key, sizeof(c->key),
 149                                        NULL, 0,
 150                                        program, program_size,
 151                                        &c->prog_data,
 152                                        &brw->wm.prog_data );
 153 }
 154
 155
 156
 157 static void brw_wm_populate_key( struct brw_context *brw,
 158                                  struct brw_wm_prog_key *key )
 159 {
 160    /* BRW_NEW_FRAGMENT_PROGRAM */
 161    struct brw_fragment_program *fp =
 162       (struct brw_fragment_program *)brw->fragment_program;
 163    GLuint lookup = 0;
 164    GLuint line_aa;
 165    GLuint i;
 166
 167    memset(key, 0, sizeof(*key));
 168
 169    /* Build the index for table lookup
 170     */
 171    /* _NEW_COLOR */
 172    if (fp->program.UsesKill ||
 173        brw->attribs.Color->AlphaEnabled)
 174       lookup |= IZ_PS_KILL_ALPHATEST_BIT;
 175
 176    if (fp->program.Base.OutputsWritten & (1<<FRAG_RESULT_DEPR))
 177       lookup |= IZ_PS_COMPUTES_DEPTH_BIT;
 178
 179    /* _NEW_DEPTH */
 180    if (brw->attribs.Depth->Test)
 181       lookup |= IZ_DEPTH_TEST_ENABLE_BIT;
 182
 183    if (brw->attribs.Depth->Test &&
 184        brw->attribs.Depth->Mask) /* ?? */
 185       lookup |= IZ_DEPTH_WRITE_ENABLE_BIT;
 186
 187    /* _NEW_STENCIL */
 188    if (brw->attribs.Stencil->Enabled) {
 189       lookup |= IZ_STENCIL_TEST_ENABLE_BIT;
 190
 191       if (brw->attribs.Stencil->WriteMask[0] ||
 192           brw->attribs.Stencil->WriteMask[brw->attribs.Stencil->_BackFace])
 193          lookup |= IZ_STENCIL_WRITE_ENABLE_BIT;
 194    }
 195
 196    line_aa = AA_NEVER;
 197
 198    /* _NEW_LINE, _NEW_POLYGON, BRW_NEW_REDUCED_PRIMITIVE */
 199    if (brw->attribs.Line->SmoothFlag) {
 200       if (brw->intel.reduced_primitive == GL_LINES) {
 201          line_aa = AA_ALWAYS;
 202       }
 203       else if (brw->intel.reduced_primitive == GL_TRIANGLES) {
 204          if (brw->attribs.Polygon->FrontMode == GL_LINE) {
 205             line_aa = AA_SOMETIMES;
 206
 207             if (brw->attribs.Polygon->BackMode == GL_LINE ||
 208                 (brw->attribs.Polygon->CullFlag &&
 209                  brw->attribs.Polygon->CullFaceMode == GL_BACK))
 210                line_aa = AA_ALWAYS;
 211          }
 212          else if (brw->attribs.Polygon->BackMode == GL_LINE) {
 213             line_aa = AA_SOMETIMES;
 214
 215             if ((brw->attribs.Polygon->CullFlag &&
 216                  brw->attribs.Polygon->CullFaceMode == GL_FRONT))
 217                line_aa = AA_ALWAYS;
 218          }
 219       }
 220    }
 221
 222    brw_wm_lookup_iz(line_aa,
 223                     lookup,
 224                     key);
 225
 226
 227    /* BRW_NEW_WM_INPUT_DIMENSIONS */
 228    key->projtex_mask = brw->wm.input_size_masks[4-1] >> (FRAG_ATTRIB_TEX0 - FRAG_ATTRIB_WPOS);
 229
 230    /* _NEW_LIGHT */
 231    key->flat_shade = (brw->attribs.Light->ShadeModel == GL_FLAT);
 232
 233    /* _NEW_TEXTURE */
 234    for (i = 0; i < BRW_MAX_TEX_UNIT; i++) {
 235       const struct gl_texture_unit *unit = &brw->attribs.Texture->Unit[i];
 236       const struct gl_texture_object *t = unit->_Current;
 237
 238       if (unit->_ReallyEnabled) {
 239          if (t->Image[0][t->BaseLevel]->InternalFormat == GL_YCBCR_MESA) {
 240             key->yuvtex_mask |= 1<<i;
 241             if (t->Image[0][t->BaseLevel]->TexFormat->MesaFormat ==
 242                     MESA_FORMAT_YCBCR)
 243                 key->yuvtex_swap_mask |= 1<< i;
 244          }
 245       }
 246    }
 247
 248    /* Shadow */
 249    key->shadowtex_mask = fp->program.Base.ShadowSamplers;
 250
 251    /* _NEW_BUFFERS */
 252    /*
 253     * Include the draw buffer origin and height so that we can calculate
 254     * fragment position values relative to the bottom left of the drawable,
 255     * from the incoming screen origin relative position we get as part of our
 256     * payload.
 257     *
 258     * We could avoid recompiling by including this as a constant referenced by
 259     * our program, but if we were to do that it would also be nice to handle
 260     * getting that constant updated at batchbuffer submit time (when we
 261     * hold the lock and know where the buffer really is) rather than at emit
 262     * time when we don't hold the lock and are just guessing.  We could also
 263     * just avoid using this as key data if the program doesn't use
 264     * fragment.position.
 265     *
 266     * This pretty much becomes moot with DRI2 and redirected buffers anyway,
 267     * as our origins will always be zero then.
 268     */
 269    if (brw->intel.driDrawable != NULL) {
 270       key->origin_x = brw->intel.driDrawable->x;
 271       key->origin_y = brw->intel.driDrawable->y;
 272       key->drawable_height = brw->intel.driDrawable->h;
 273    }
 274
 275    /* Extra info:
 276     */
 277    key->program_string_id = fp->id;
 278
 279 }
 280
 281
 282 static void brw_prepare_wm_prog(struct brw_context *brw)
 283 {
 284    struct brw_wm_prog_key key;
 285    struct brw_fragment_program *fp = (struct brw_fragment_program *)
 286       brw->fragment_program;
 287
 288    brw_wm_populate_key(brw, &key);
 289
 290    /* Make an early check for the key.
 291     */
 292    dri_bo_unreference(brw->wm.prog_bo);
 293    brw->wm.prog_bo = brw_search_cache(&brw->cache, BRW_WM_PROG,
 294                                       &key, sizeof(key),
 295                                       NULL, 0,
 296                                       &brw->wm.prog_data);
 297    if (brw->wm.prog_bo == NULL)
 298       do_wm_prog(brw, fp, &key);
 299 }
 300
 301
 302 /* See brw_wm.c:
 303  */
 304 const struct brw_tracked_state brw_wm_prog = {
 305    .dirty = {
 306       .mesa  = (_NEW_COLOR |
 307                 _NEW_DEPTH |
 308                 _NEW_STENCIL |
 309                 _NEW_POLYGON |
 310                 _NEW_LINE |
 311                 _NEW_LIGHT |
 312                 _NEW_BUFFERS |
 313                 _NEW_TEXTURE),
 314       .brw   = (BRW_NEW_FRAGMENT_PROGRAM |
 315                 BRW_NEW_WM_INPUT_DIMENSIONS |
 316                 BRW_NEW_REDUCED_PRIMITIVE),
 317       .cache = 0
 318    },
 319    .prepare = brw_prepare_wm_prog
 320 };
 321